Drone Plot Harvester
Sept 2016 Update - Plot harvest and efficacy videos
April 2016 First Field Trial - Trial video
The Drone Plot Harvester has been developed over the last few years with many design iterations and it has generated a fair sized scrap heap as well. It's now at first working prototype stage. It is one of the few operational harvest drones in the world for any crop.
At this point it needs an operator in the loop, controlling the drone. The plan is to gradually add self-driving functionality so that the operator can work on the loop, directing two to four semi-autonomous units. Eventually this would be enhanced to the point where the operator is out of the loop, monitoring an unspecified number of fully autonomous drones, possibly in remote locations.
As it is now it has a number of unique features
Very
low impact on soils
The drone has an overall weight of about 75Kg fully laden with batteries and nuts for a 750mm wide pickup. There are no land wheels to concentrate the weight and it is self propelled obviating the need for a large and heavy prime mover to push it. The result is negligible soil compaction and the ability to work in wetter conditions, for example sooner than is usual after rain.
Fully reversible with direct driven fingerwheels
The design is actually completely symmetrical, it doesn't have a forward and reverse so much as simply directions 'A' and 'B' and it will pick up equally well in either mode. This saves time and batteries by reducing the amount of turning required. The constant changing of direction also stops trash and snags building up on the wheels.
The wheels are direct driven as opposed to the normal ground-friction drive and thus do not jam in slippery conditions.
Highly maneuverable
The differential drive allows the machine to turn on itself while still harvesting nuts so it can harvest even in the most awkward to access areas. The tree line can be harvested without the need for blowers or sweepers, eliminating a significant cause of soil degradation. The thoroughness of the pickup can be adjusted tree by tree depending on how heavy the crop is.
In addition to these points it has low running costs, potential for fully autonomous operation and the ability to create tree by tree yield maps at minimal additional cost.
The harvester is currently aimed at trial plots.
For those not familiar with macadamia yield trial work, macadamias fall to the ground when they are mature and from there they get picked up. Commercially this is usually done with 'fingerwheel' type harvesters that have a gang of fingerwheels that entrap the nuts as they get run over and then extracts them into a transport system using combs (see images). However while these machines are effective they are incapable of accurately stop-starting in order that tree-by-tree yields can be retrieved. Thus all research plots are harvested by hand which is expensive and hard work. There are hand push versions of fingerwheel type harvesters but they are only moderately effective and it is difficult to get them under the trees when the canopy skirt is low.
This drone also uses the fingerwheel system under the hood but it aims for a point in the middle of the three options that is relatively easy, effective and efficient. We use it for plot harvesting in tandem with the in-field dehusker so one tree is dehusked as the next one is harvested. At each changeover the nut in shell of the previous tree is weighed and can be quickly assessed for a number of traits. In this way it is possible for one person to comfortably harvest and process a tree roughly every two to three minutes. For high value trials it would be better to have a second person running the dehusker and recording data to reduce the chance of mix ups and to ensure samples are clean when weighed.
We are attempting to use the drone for our main commercial harvest this year, it's the only way to see how well it works for this purpose and to find all the design faults before a final release version is built. If a fully autonomous system is to be developed a prerequisite will be a unit that is completely reliable mechanically.
While it is slower than a normal commercial systems we hope that its ability to work in conditions where other harvesters stay in the shed, to concurrently do post-harvest steps and to constantly adjust for conditions on a tree by tree basis will allow it to have an overall efficiency similar to normal harvesters. A bit like The Hare & The Tortoise really, check back in twelve months to see how that went :-)
As higher levels of autonomous operation such as self-driving capability
are gradually added to the system it will become an increasingly
compelling option for commercial operations.
As mentioned above the harvester will be tested on our main commercial crop this year (2016). Depending on how well that goes the plan would be to use the lessons from this season to produce a new Release Design next season (2017) though realistically that is unlikely before the end of that season. Interested parties should contact me via the panel on the left.
Mapping and estimating location within maps is a critical prerequisite for autonomous navigation.
This video shows the orchard mapping system in action, it actually uses an early prototype of the sense hardware that has since been scrapped. However I thought I would put it up for reference as I hope to have some videos of the new hardware in the next few weeks.
The system in the video uses
It works quite well but there are number of issues in the output maps if you look closely, so a few improvements are being implemented in the new hardware.
Videos first ....
This video shows three examples of performing a plot harvest operation in
~90 seconds/plot.
The original field trial video is further down and it's interesting to see the progress during the last few months.
This video shows a few examples of the pickup efficiency. While the
harvester is based on tried and true fingerwheel pickup method the direct
drive and extra ground pressure improves the efficacy.
Results for 2016 season
Changes made since April
Overall the season has gone well; most of the issues identified in the April trial have been resolved. One of the main goals for this season was to refine the drone to the point where it has very high mechanical reliability - a prerequisite for looking into autonomy. This has largely been achieved, it essentially has 100% 'up time' and the time between problems that need unplanned manual intervention is getting up into weeks now.
We are on track for the initial goal of doing our whole commercial harvest with the drone, over ten tonnes have been harvested it to date. The drone had a number of issues at the start of the season that made it too slow but these were largely resolved by early August. As it stands the harvest will be finished later than normal but if the unit was in its current state back in April we would have easily finished on time.
It currently harvests at about 100kg/hr in fairly poor conditions but moderately heavy crop. The nuts are also fully dehusked in field in the same process. With the planned enhancements outlined below this should come up to ~200kg/hr for the start of next season.
Positives:
Negatives:
Upgrading the dehusker to a complete in-field post harvest processing
system will be the major focus of the 2017 season and will be documented
in due course on the Dehusker page to the left.
However there are a number of changes planned for the drone as well.
In additional to the above there are a couple of significant changes to the drone unit planned if there is enough time to get them done.
Initial results of the first full field trial of the drone
Positives:
Negatives:
So in conclusion the biggest change to come out of this trial is that the motors will be upgraded. There might also be some software changes, adjustment of clearances etc.
In the video below I am still something of an 'L-plate' driver. About half way through I thought it had died but it turned out I had just pressed the wrong button on my control panel.
Copyright © 2016 David Bell, Australia. All Rights
Reserved.
No parts of this web site may be copied without the
permission of the author.